Experimental codes are implemented in R. For codes to work properly, set the working directory as below:
setwd("~/Mixed-categorical-ordered-imputation-extended-Gaussian-copula/Categorical_EGC")
with ~ replaced by the relative path in your place.
The imputation algorithm of the extended Gaussian copula is implemented in func_EGC.R. Its implementation depends on the R package gcimputeR and rootSolve. rootSolve can be installed from CRAN. gcimputeR can be installed from Github:
library(devtools)
install_github("udellgroup/gcimputeR")
The following R packages are also required for experiments: missForest, missMDA, softImpute, mice, purrr. All of them can be installed directly from CRAN.
As discussed in Sec 2.1 of the supplement, EGC uses a Python implementation to estimate the marginal and a R implementation to estimate the copula correlation. In this repo, we use a complete R implementation of EGC for simplicity, which is slower. A complete Python implementation of EGC will be available soon.
Main results can be replicated using file main_sim.R and main_realdata.R.